A noninvasive ECG-based machine-learning tool for early detection of left atrial appendage thrombus in first-diagnosed atrial fibrillation

Background The timely identification of left atrial appendage thrombus (LAAT) is paramount for stroke prevention in non-valvular atrial fibrillation (NVAF), yet current reliance on transesophageal echocardiography (TEE) limits accessibility. Objectives To develop and validate a non-invasive, electrocardiogram (ECG)-based machine-learning model for predicting LAAT in patients with first-diagnosed NVAF. Methods A retrospective cohort of 486 patients who underwent ECG within 2 days before TEE or cardiac CT was analyzed. Eight predictive ECG features were selected using Lasso regression. Five machine-learning models were trained and evaluated. Model performance was assessed by the area under the receiver operating characteristic curve (AUC), calibration, and decision curve analysis (DCA). Continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were used to compare against the CHA 2 DS 2 –VASc score. Results LAAT was present in 87 patients (17.9%). The logistic regression (LR) model demonstrated optimal performance, achieving an AUC of 0.803 (95% CI 0.750–0.855), sensitivity of 0.77, specificity of 0.81, and excellent calibration (Hosmer–Lemeshow χ 2 = 7.295, df = 8, P = 0.505). DCA showed net clinical benefit across thresholds of 4–75%. It significantly outperformed CHA 2 DS 2 –VASc (AUC = 0.570), yielding a continuous NRI of 0.9228 and an IDI of 0.1882. Adding CHA 2 DS 2 –VASc did not improve prediction. The final model was deployed as an interactive online calculator for real-time risk assessment. Conclusions The ECG-based logistic regression model provides an interpretable approach with promising internally validated performance for predicting LAAT in NVAF. This tool enables rapid, non-invasive risk stratification and may help prioritize patients who warrant expedited definitive imaging; its clinical utility, including any role in anticoagulation decisions, requires external prospective validation.